Predicting Falls in Older Adults using AI-Powered Wearable IoT Devices: A Scoping Review

Abstract

Background: Falls among older adults are a major global health concern associated with disability, hospitalization, mortality, and reduced independence. Recent advances in artificial intelligence (AI), wearable Internet of Things (IoT), and machine learning have shifted fall prevention from reactive detection toward proactive risk prediction. Objective: This scoping review mapped current evidence on AI-powered wearable IoT devices for predicting falls in older adults, focusing on sensor technologies, AI models, predictive performance, and implementation challenges. Methods: Following PRISMA-ScR guidelines, literature searches were conducted in PubMed, Scopus, IEEE Xplore, Web of Science, and Cochrane Library databases for studies published between 2008 and 2026. Eligible studies involved older adults, wearable IoT sensors, and AI or machine learning models for fall-risk prediction or pre-impact fall detection. Results: Of 1,100 identified records, 170 full-text articles were screened, and 13 studies met the final eligibility criteria. IMUs, accelerometers, gyroscopes, plantar-pressure sensors, and surface electromyography (sEMG) were the most frequently used sensors. Random Forest, XGBoost, Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) were the dominant analytical approaches. Several laboratory-based systems achieved accuracies above 95%; however, most studies relied on simulated falls in younger individuals rather than on real-world data from older adults. Conclusion: AI-powered wearable IoT systems show strong potential for proactive fall-risk prediction in older adults. However, translational barriers remain, including limited ecological validation, small sample sizes, and the persistent simulation gap. Future research should prioritize real-world validation, explainable AI, and the integration of multimodal sensing.